A Multi-Distance Ensemble of Multi-Criteria Decision Making for Ontology Ranking
Ameeth Sooklall, Jean Vincent Fonou-Dombeu · Future Internet · 2026
Due to the increase in the number of ontologies in various domains, ranking them to facilitate their selection for reuse is an important task in ontology engineering to date. To assess the multi-faceted quality configurations of candidate ontologies, Multi-Criteria Decision Making (MCDM) frameworks are used. In particular, the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) is an MCDM method that is widely adopted for the task of ontology ranking. However, traditional TOPSIS implementations rely almost exclusively on the Euclidean distance metric. This introduces severe rank volatilities and systematic biases when evaluating heterogeneous ontology metadata. To address these limitations, this paper introduces a novel Multi-Distance Ensemble TOPSIS (Ensemble-TOPSIS) method for robust ontology ranking. Rather than forcing a localized geometric choice, the proposed Ensemble-TOPSIS method simultaneously projects alternative ontologies through a multi-distance ensemble composed of Euclidean, Chebyshev, cosine, and Mahalanobis configurations. The Ensemble-TOPSIS method was applied to three datasets of ontologies from the artificial intelligence, agricultural, and biological domains to test its scalability and multi-domain applicability. The experimental results reveal that all the ontologies from the three domains were successfully ranked by the proposed Ensemble-TOPSIS method. Furthermore, the statistical rank correlation using Spearman’s ρ, Kendall’s τ, and the WS rank similarity coefficients was calculated between the TOPSIS variants, and the proposed Ensemble-TOPSIS method achieved the highest correlation in the majority of cases. Moreover, a comprehensive Monte Carlo simulation across 1200 stochastically generated, non-linear, and skewed multicollinear decision domains established the asymptotic stability of the proposed Ensemble-TOPSIS method, which achieved the highest global mean performance (ρ¯=0.88, τ¯=0.75, WS¯=0.94), minimized rank variance (σ2(WS)=0.0006), and optimally maximized the lower-bound worst-case performance profile (ρ=0.67) compared to individual baseline formulations.